ktx/packages/cli/src/context/llm/runtime-port.ts
Andrey Avtomonov 21744fc520
feat(cli): profile ingest runs and split model vs tool time (#249)
* feat(cli): profile ingest runs to find where wall-clock time goes

Add opt-in profiling for `ktx ingest`. Each timed phase, work unit, and
agent loop now records durationMs / step count / token usage in the
trace, and a post-run aggregator rolls them up into a "where did the
time go" report printed to stderr.

Enable per run with KTX_PROFILE_INGEST (1/true -> human table, json ->
raw structured profile) or persistently via `ingest.profile` in
ktx.yaml. The json form emits raw milliseconds, token counts, and a
summary.headline one-line diagnosis so coding agents can parse it
directly; json wins when both env and config request profiling.

- runtime-port: RunLoopMetrics (totalMs, usage, stepCount,
  stepBoundariesMs) plus onMetrics callbacks on text/object generation
- ai-sdk + claude-code runtimes: capture per-loop timing and token usage
- work-unit-executor and stages 3/4: thread metrics into trace events
- ingest-bundle.runner: time worktree / triage / clustering / index /
  reconcile / squash phases and emit the profile in a finally block
  (best-effort; never affects the run outcome)
- ingest-profile: new trace+transcript aggregator with table/json formatters
- config: ingest.profile flag; docs: profiling section in ktx-ingest.mdx

* fix(cli): flush tool-call logs before reading ingest profile

Tool transcripts are appended fire-and-forget so the agent hot path never
blocks on logging. The ingest profiler read them before the writes settled,
so per-work-unit toolMs (and the model-vs-tool split derived from it) could
be incomplete. Track in-flight appends and expose flushToolCallLogs() —
bounded by a timeout so it can never hang — and flush before the profiler
reads the transcript.
2026-06-01 15:49:17 +02:00

97 lines
2.7 KiB
TypeScript

import type { KtxModelRole } from '../../llm/types.js';
import type { z } from 'zod';
export interface KtxRuntimeToolOutput<TOutput = unknown> {
markdown: string;
structured?: TOutput;
}
export interface KtxRuntimeToolDescriptor<TInput = unknown, TOutput = unknown> {
name: string;
description: string;
inputSchema: z.ZodObject<z.ZodRawShape>;
execute(input: TInput): Promise<KtxRuntimeToolOutput<TOutput>>;
}
export type KtxRuntimeToolSet = Record<string, KtxRuntimeToolDescriptor>;
export type RunLoopStopReason = 'budget' | 'natural' | 'error';
/** @internal */
export interface RunLoopStepInfo {
stepIndex: number;
stepBudget: number;
}
export interface LlmTokenUsage {
inputTokens?: number;
outputTokens?: number;
totalTokens?: number;
}
/** Timing and token metrics for a multi-step agent loop, used for ingest profiling. */
export interface RunLoopMetrics {
/** Wall-clock time around the whole `generateText` call, in milliseconds. */
totalMs: number;
/** Aggregate token usage across all steps. */
usage: LlmTokenUsage;
/** Number of agent steps (model round-trips) that actually ran. */
stepCount: number;
/** Wall-clock offset (ms from loop start) at which each step finished. */
stepBoundariesMs: number[];
}
export interface RunLoopParams {
modelRole: KtxModelRole;
systemPrompt: string;
userPrompt: string;
toolSet: KtxRuntimeToolSet;
stepBudget: number;
telemetryTags: Record<string, string>;
onStepFinish?: (info: RunLoopStepInfo) => void | Promise<void>;
}
export interface RunLoopResult {
stopReason: RunLoopStopReason;
error?: Error;
metrics?: RunLoopMetrics;
}
export interface KtxGenerateTextInput {
role: KtxModelRole;
prompt: string;
system?: string;
tools?: KtxRuntimeToolSet;
temperature?: number;
onMetrics?: (metrics: { totalMs: number; usage: LlmTokenUsage }) => void;
}
export interface KtxGenerateObjectInput<TOutput, TSchema extends z.ZodType<TOutput>> {
role: KtxModelRole;
prompt: string;
system?: string;
tools?: KtxRuntimeToolSet;
temperature?: number;
schema: TSchema;
onMetrics?: (metrics: { totalMs: number; usage: LlmTokenUsage }) => void;
}
export interface KtxLlmRuntimePort {
generateText(input: KtxGenerateTextInput): Promise<string>;
generateObject<TOutput, TSchema extends z.ZodType<TOutput>>(
input: KtxGenerateObjectInput<TOutput, TSchema>,
): Promise<TOutput>;
runAgentLoop(params: RunLoopParams): Promise<RunLoopResult>;
}
export interface AgentRunnerPort {
runLoop(params: RunLoopParams): Promise<RunLoopResult>;
}
export class RuntimeAgentRunner implements AgentRunnerPort {
constructor(private readonly runtime: KtxLlmRuntimePort) {}
runLoop(params: RunLoopParams): Promise<RunLoopResult> {
return this.runtime.runAgentLoop(params);
}
}